Finding Water Loss with Better Data

When a utility produces more water than its billing records appear to explain, the gap can look like a straightforward technology problem: find the leaking pipes.

Sometimes that is part of the answer. The gap can also involve meter error, differences between reporting periods, authorized uses that are not billed, or incomplete records. A leak-detection product cannot correct every kind of discrepancy.

Better data is valuable when it helps a utility distinguish these problems and choose a proportionate next step. The starting point is a water balance, not a map filled with alerts.

Separate the kinds of loss

The EPA’s water-audit guidance distinguishes real losses from apparent losses. Real losses involve physical leakage or overflow. Apparent losses involve water consumed but not correctly accounted for because of factors such as metering inaccuracies, unauthorized consumption, and data-handling errors.

Authorized consumption also matters. Some legitimate uses may not be billed, so non-revenue water is not simply another term for leakage.

Those distinctions lead to different interventions. A damaged pipe may require field work. A meter problem may require testing and replacement. A billing error may require correcting records and processes.

Do not let a single “water loss” number hide the mix. The estimate is useful only when the team knows how its components were assembled and how confident it can be in them.

Make the accounting periods agree

Start with the boundary of the system and a common period. Water entering a distribution system and water recorded at customer meters must describe compatible quantities.

Record how production meters, customer reads, exports, imports, storage changes, and authorized uses enter the calculation. Check units. Preserve the distinction between a measured value, an estimate, and a missing record.

For an illustrative example, a production figure covering a calendar month and customer readings taken on staggered schedules may not be directly comparable. Treating their difference as a precisely measured leak volume would overstate what the records establish.

The answer is not to abandon the audit. It is to document the method and improve the uncertain inputs.

The EPA’s distribution-system reference places water auditing at the beginning of loss management. The audit directs investigation; it does not eliminate the need for it.

Match the tool to the uncertainty

Different tools answer different questions. A utility should identify which uncertainty blocks a decision before choosing another measurement.

These are complementary approaches, not a universal order of purchase. A small utility’s most valuable first intervention may be a corrected meter inventory. Another may already have adequate accounting and need targeted field investigation.

Treat a supplier’s claimed detection rate as a question to evaluate under local conditions. Ask which types of leak, materials, installation environments, and reference checks the claim covers.

Use patterns without confusing them with proof

A change in overnight flow can be informative. It can also reflect legitimate uses, an operating change, or a measurement problem. A useful alert narrows an investigation; it does not automatically identify a failed pipe.

Record what information an operator uses to judge an alert. A time series becomes easier to interpret when it is connected to maintenance events, unusual demand, valve work, meter replacement, and changes to a zone boundary.

Avoid hiding gaps through smoothing. An attractive continuous chart may conceal that a device failed during the period the team needs to understand.

When sensor systems are proposed, evaluate the full measurement chain: installation, communications, timestamps, validation, interpretation, and response.

Compare performance with care

A percentage is easy to communicate but can be difficult to interpret. Changes in legitimate demand can change a percentage even when physical losses are unchanged. Different system sizes and operating conditions also complicate comparisons.

Use indicators appropriate to the local purpose and supported by the audit method. Document the denominator and the period. Where a comparison informs an engineering or investment decision, involve the utility’s qualified technical staff.

Keep physical volume, financial implications, and resource value distinct. Correcting an under-recording meter can improve accounting without repairing a pipe. Repairing leakage can reduce wasted treated water without producing an identical change in billed revenue.

EPA’s conservation and efficiency guidance places loss management in a wider resource-planning context. For a city, that means asking about capacity and operational value as well as a short-term bill.

Budget for the response

A better detection system creates work. Someone must validate alerts, arrange access, plan repairs, update records, and verify the result.

If that capacity is absent, a larger alert queue may add little immediate public value. Plan the operating process alongside the data collection.

For each investigation, preserve a short record: the signal, the evidence checked, the finding, the action, and the follow-up observation. Include false leads. They help reveal whether thresholds or measurement locations need to change.

This log also prevents a common overclaim: attributing every reduction in a system-level figure to a single new tool when several operational changes occurred at once.

Start with a bounded question

An achievable first project might be reconciling a defined set of meters, improving one uncertain audit input, or investigating a clearly bounded zone. It should have an owner, a method, and a decision it can inform.

Before expanding, ask whether the new information changed an action, whether the result could be verified, and whether the team can sustain the work.

The discipline resembles building-energy analysis: establish a trustworthy boundary, interpret the record, and investigate before asserting a cause.

A city does not need perfect data to begin reducing uncertainty. It does need to know which part of the picture is measured, which is estimated, and which remains unexplained.